Segmentation and modeling of MRI image

Development
1

Hi,

I want to implement the segmentation of bladder from pelvic floor MRI images and then implement 3D reconstruction of the organ in a python scripted module.
I currently created a model which can predict by each pixel of the MRI image. The model returns a (227,227,30) matrix of value 0 and 1. (the MRI image has 30 images with 227*227 pixels). In the matrix, 1 indicates this pixel is a part of bladder and 0 is not. This model tells the position of the segmented organ in the image.
For the 3D Modeling part, I think the Show 3D and Fill between slices of the Segment Editor module will do it. But how can I segment the organ by the matrix predicted by my model? Does anyone have any idea how to implement this?

Thanks for your time!
Andrea

2

The data you call model we call binary labelmap. Please load is to Slicer as a labelmap (choose labelmap in the options of the load window, or convert it to labelmap after loading in the Volumes module), then import it as a segmentation (right-click labelmap in Data module and choose import labelmap to segmentation).

3

Hi, thanks so much for your help.
Your idea seems practical but Iā€™m sorry I didnā€™t completely understand. I load MRI images in the format .gz to Slicer, and the model is a .h5 file which is made of a convolutional neural network. My model can predict the MRI images to a (227,227,30) matrix. What should I load to Slicer? Iā€™m new to Slicer so sorry for lack of knowledge.

4

Save the ā€œmodelā€ (we call it labelmap volume) as a .nrrd or .mha file and you will be able to load it as segmentation or labelmap volume.

5

Hi, thanks for answering.
Iā€™m a little confused. The ā€œmodelā€ is a deep convolutional neural network, how can I load it as volume?

6

@ZiyunLiang, what you call the ā€œmatrixā€ (output of your neural network ā€œmodelā€) is what we call a labelmap, so if you load that into Slicer, you can make it into a what Slicer calls a ā€œmodelā€ which is a triangulated surface. :grinning:

1 Like
7

Thatā€™s right. matrix = labelmap, not model as I said.

You cannot import your model to Slicer, but if itā€™s python or C++ or Matlab, then you can integrate it within a new module.

1 Like
8

Hi, really appreciate your help and explanation, I have figured out a way to implement this now. However, I still got a problem when Iā€™m trying to write it in my scripted python module.
When Iā€™m trying to convert my volume to labelmap by using Volumes module, Volume information section, and Convert to label map
I used the code

volumeNode=slicer.util.getNode(ā€˜vtkMRMLScalarVolumeNode1ā€™)
labelmap=slicer.mrmlScene.AddNewNodeByClass('vtkMRMLLabelMapVolumeNodeā€™)
slicer.modules.Volumes.logic().CreateLabelVolumeFromVolume(vtkMRMLScene,label map,volumeNode)

But there is something wrong.

AttributeError: ā€˜moduleā€™ object has no attribute ā€˜Volumesā€™

I donā€™t have much experience in writing python code in 3d Slicer so please help me out. What is the correct way of realizing this?

Thanks for your time!

9

You have the case wrong which is causing the attribute error.

slicer.modules.Volumes.logic().CreateLabelVolumeFromVolume(vtkMRMLScene,label map,volumeNode)

should be

slicer.modules.volumes.logic().CreateLabelVolumeFromVolume(vtkMRMLScene,label map,volumeNode)

A tip is also to press {TAB} after any period to see the available options in an auto-complete type of way.

10

Hi, thanks for the quick response and the tip, Iā€™ll keep that in mind. I tried this but I think Iā€™m still implementing this function wrong. What should the vtkMRMLScene be?

NameError: name ā€˜vtkMRMLSceneā€™ is not defined

11

Yes any variable needs to be defined as well. In this case, vtkMRMLScene wasnā€™t defined. We can instead use ā€œslicer.mrmlSceneā€ which is defined.

slicer.modules.volumes.logic().CreateLabelVolumeFromVolume(slicer.mrmlScene, labelmap, volumeNode)
12

Hi James,
Thanks so much for your help. Itā€™s working now.:grinning:

1 Like
13

Instead of this approach, Iā€™d load the volume as labelmap to begin with:

slicer.util.loadLabelVolume('d:/path/to/Voulme.nrrd',{})

14

This is indeed an easier way to do this. Thanks:smiley: